Probabilistic programming with stochastic variational message passing

Stochastic approximation methods for variational inference have recently gained popularity in the probabilistic programming community since these are amenable to automation and allow online, scalable, universal approximate Bayesian inference. Unfortunately, common Probabilistic Programming Languages (PPLs) with stochastic engines lack efficiency of message passing-based algorithms deterministic update rules such as Belief Propagation (BP) Variational Message Passing (VMP). Still, Inference (SVI) Conjugate-Computation (CVI) provide principled integrate fast techniques broadly applicable implementation SVI CVI necessitates manually driven rules, which does not yet exist most PPLs. In this paper, we cast explicitly a context. We an ForneyLab, is automated package open source Julia language. Through number experiments, demonstrate how extends capabilities programming.